Jet Calibration and Mixture Density Networks

The energy calibration for jets is important for many physics measurements at the Large Hadron Collider. Jets are sprays of particles in the detector originating from quarks and gluons. The ATLAS detector is composed of the inner charged particle tracking detector, electromagnetic and hadronic calorimeters, and a muon spectrometer. The information gained from these layers contributes to the process of jet reconstruction. This process is complicated by the large number of overlapping collisions known as pileup. The ATLAS detector will be upgraded for the High Luminosity Large Hadron Collider (HL-LHC) where the pileup will be much higher. An important step in the jet reconstruction is the Monte Carlo based calibration (MCJES) which corrects for overall jet energy scale. There is ongoing effort to replace this step with a machine learning (ML) regression that quickly learns and performs improved calibrations. In this talk, I will present a possible method using a mixture density network, a model that draws its predictions from a constructed probability density function, to perform a ML-based MCJES calibration for Run 3 and Simulated HL-LHC samples. There will be an overview of mixture density networks, performance compared to other deep learning networks, and future plans for the project.